Other Roles Matter! Enhancing Role-Oriented Dialogue Summarization via Role Interactions

Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 2545 - 2558 May 22-27, 2022 c⃝2022 Association for Computational Linguistics Other Roles Matter! Enhancing Role-Oriented Dialogue Summarization via Role Interactions Haitao Lin1,2, Junnan Zhu1,2, Lu Xiang1,2, Yu Zhou1,3∗, Jiajun Zhang1,2, Chengqing Zong1,2 1 National Laboratory of Pattern Recognition, Institute of Automation, CAS, Beijing, China 2 School of Artificial Intelligence, University of Chinese Academy of Sciences, Beijing, China 3 Fanyu AI Laboratory, Zhongke Fanyu Technology Co., Ltd, Beijing, China {haitao.lin, junnan.zhu, lu.xiang, yzhou, jjzhang, cqzong}@nlpr.ia.ac.cn, Abstract Role-oriented dialogue summarization is to generate summaries for different roles in the dialogue, e.g., merchants and consumers. Ex- isting methods handle this task by summariz- ing each role’s content separately and thus are prone to ignore the information from other roles. However, we believe that other roles’ content could benefit the quality of summaries, such as the omitted information mentioned by other roles. Therefore, we propose a novel role interaction enhanced method for role-oriented dialogue summarization. It adopts cross atten- tion and decoder self-attention interactions to interactively acquire other roles’ critical infor- mation. The cross attention interaction aims to select other roles’ critical dialogue utter- ances, while the decoder self-attention inter- action aims to obtain key information from other roles’ summaries. Experimental results have shown that our proposed method signif- icantly outperforms strong baselines on two public role-oriented dialogue summarization datasets. Extensive analyses have demonstrated that other roles’ content could help generate summaries with more complete semantics and correct topic structures.1 1 Introduction Dialogue summarization aims at compressing the main content of a long conversation into a short text. With the development of online conversation tools, the amount and length of conversation are growing up rapidly. Since a dialogue often contains compli- cated structure and ellipsis, it is time-consuming to read the whole dialogue. Dialogue summarization thus becomes valuable since it could extract the key point of a conversation and greatly reduce the time cost. This technique is widely used in customer service (Liu et al., 2019), meeting (McCowan et al., 2005), online chatting (Gliwa et al., 2019), etc. ∗Corresponding author. 1Our codes are available at: https://github.com/ xiaolinAndy/RODS. 0 Q: ������������������������� (My original phone number was useless, and I forgot my password. ) 1 A: �������������� (Can I transfer you to the commissioner here?) 2 Q: ���(Yes.) 3 A: �������������������[��]��������� (I'm sorry, the transfer failed. Can I help to upgrade the commissioner and call you back in [NUM] hours?) 4 Q: ������(It's OK.) 5 A: ��������������? (Is there anything else I can help you with?) 6 Q: ���� ������������ (By the way, Can JD pay via wechat? ) 7 A: ���������(Yes, it is OK normally.) Dialogue ������������������������������ ��� ������(The customer service helped the user to call back within [number] hours after the transfer specialist fails. The customer service said JD could pay via wechat.) Agent Summary User Summary : Information from the user’s utterances, could enhance agent summary in : Information from the agent’s summaries, could enhance …

Content truncated to 500 words